DeepAnnotation: A novel interpretable deep learning-based genomic selection model that integrates comprehensive...

Area of Science:

• Agricultural Science
• Bioinformatics
• Genetics

Background:

• Genomic selection accelerates livestock trait improvement using genomic information.
• Multiomics data offers potential to enhance genomic selection with biological knowledge.
• Accurate phenotype prediction using multiomics data in genomic selection is still developing.

Purpose of the Study:

• To develop DeepAnnotation, an interpretable deep learning model for phenotype prediction.
• To integrate multiomics functional annotations into a deep learning framework for genomic selection.
• To improve the accuracy and efficiency of predicting economically important traits in livestock.

Main Methods:

• Developed DeepAnnotation, a deep learning model aligning multiomics annotations with sequential network layers.
• Modeled the genotype-to-phenotype cascade from cis-regulatory elements to genes, modules, and traits.
• Evaluated model performance against seven classical genomic selection models.

Main Results:

• DeepAnnotation demonstrated significantly superior prediction accuracy (6.4%–120.0% increase) compared to seven classical models.
• Achieved high computational efficiency in predicting three pork production traits (lean meat percentage, loin muscle depth, back fat thickness).
• Enabled identification of potential causal single-nucleotide polymorphisms (SNPs) and their molecular mechanisms.

Conclusions:

• DeepAnnotation is an open-source, interpretable deep learning tool for phenotype prediction in genomic selection.
• The model effectively leverages multiomics functional annotations for enhanced prediction.
• Provides a valuable resource for researchers and practitioners in livestock breeding and genetics.